Skill: close

Transition a story from done to archived with retrospective and knowledge extraction.

Inputs

• Story file: Path to a .swarm/backlog/{ID}.md file in done status

Steps

Step 1: Read Context

1. Read .swarm/config.md for project conventions.
2. Read the story file — gather all tasks, ACs, mini-retrospectives.
3. Read the WoW to check if knowledge_extraction is enabled.

Step 2: Compute Metrics

1. Count tasks: total, completed, skipped.
2. Count ACs: total, passing, failing.
3. Count files changed (via git diff --stat or similar).
4. Count tests added.
5. Calculate cycle time from execution.started_at to execution.completed_at.

Step 3: Extract Learnings per Agent

1. Parse each mini-retrospective from the story body.
2. Consolidate "What worked", "What to remember", and "Suggestion" items.
3. Group learnings by agent.

Step 4: Generate Retrospective

1. Create the retrospective file at .swarm/retrospectives/{ID}-retro.md.
2. Fill in the template with:
   • Summary of what was accomplished
   • What went well / what could improve
   • Key decisions table
   • Effort analysis table
   • Learnings by agent
   • Knowledge items extracted (with dimension, scope, and title)
3. Update the retrospective frontmatter with metrics and knowledge_extracted references.

Step 5: Extract Knowledge Items

Use the swarm_extract_knowledge tool with:

• retro_path: absolute path to .swarm/retrospectives/${STORY_ID}-retro.md
• story_id: the story ID
• repo: the repository name

This produces structured knowledge items at .swarm/knowledge/. Review the result for the list of created files.

Step 6: Update Agent Memory

1. For each agent involved, append key learnings to .claude/agent-memory/{agent}/MEMORY.md.
2. Include patterns, gotchas, and decisions relevant to that agent's role.

Step 7: Transition to Archived

Use the swarm_transition tool with path set to the absolute path of .swarm/backlog/${STORY_ID}.md and target_status set to archived.

Then move the story file:

1. Move the story file from .swarm/backlog/{ID}.md to .swarm/archive/{ID}.md.
2. Update the updated timestamp.

Step 8: List Knowledge Items

Use the swarm_list tool with directory set to the absolute path of .swarm/knowledge to verify extracted knowledge.

Outputs

• Retrospective file: .swarm/retrospectives/{ID}-retro.md
• Knowledge item files: .swarm/knowledge/{ID}-{N}.md (one per extracted item)
• Updated agent memory files
• Archived story: .swarm/archive/{ID}.md
• Story status: archived